Slope monitoring data fusion method and system based on multiple sensors

Abstract

The invention relates to the technical field of side slope monitoring, and particularly discloses a side slope monitoring data fusion method and system based on multiple sensors, and the method comprises the steps: collecting various physical parameters of a side slope through a plurality of side slope monitoring sensors, and further constructing an evidence body of each physical parameter; the basic probability distribution of each evidence body to different propositions is determined in combination with statistical analysis and expert experience, then all the evidence bodies are gradually fused by adopting a D-S evidence synthesis rule to obtain multi-sensor fusion representation, and then, based on the multi-sensor fusion representation, according to a maximum credibility decision rule, the probability distribution of each evidence body to different propositions is determined. And the final stability state of the slope is determined, and corresponding early warning measures are taken. In this way, multi-source heterogeneous data can be utilized more comprehensively, the accuracy and comprehensiveness of slope stability evaluation are improved, and a powerful guarantee is provided for engineering safety.

Description

Technical Field

[0001] This application relates to the technical field of slope monitoring, and more specifically, to a multi-sensor-based slope monitoring data fusion method and system. Background Art

[0002] The construction and operation of major projects such as highways, railways, and reservoir dams highly rely on the stability of slopes. Therefore, slope stability monitoring, as a key link to ensure project safety, is of great significance for preventing disastrous accidents. Especially in areas with complex geological conditions and frequent natural disasters, slope instability may lead to serious safety accidents and economic losses.

[0003] During slope monitoring, various types of monitoring sensors are usually used, such as displacement sensors, stress sensors, rain gauges, and radar monitoring equipment, to obtain various physical parameters and deformation information of the slope. However, the data collected by different sensors often have different accuracies, reliabilities, and uncertainties, and these data may be contradictory or complementary. Traditional data processing methods are difficult to fully utilize the advantages of these multi-source heterogeneous data and cannot effectively handle the uncertainties in the data, resulting in inaccurate and incomplete evaluation of slope stability.

[0004] Therefore, an optimized multi-sensor-based slope monitoring data fusion method and system are expected. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide a multi-sensor-based slope monitoring data fusion method and system, which use various slope monitoring sensors to collect various physical parameters of the slope, further construct evidence bodies for each physical parameter, determine the basic probability assignment of each evidence body to different propositions by combining statistical analysis and expert experience, and then use the D-S evidence synthesis rule to gradually fuse all evidence bodies to obtain a multi-sensor fusion representation. Furthermore, based on the multi-sensor fusion representation, according to the maximum confidence decision rule, the final stability state of the slope is determined and corresponding warning measures are taken. In this way, multi-source heterogeneous data can be more comprehensively utilized, the accuracy and comprehensiveness of slope stability evaluation can be improved, and strong guarantee for project safety can be provided.

[0006] Correspondingly, according to one aspect of this application, a multi-sensor-based slope monitoring data fusion method is provided, which includes:

[0007] Obtain data collected by various slope monitoring sensors, where the various slope monitoring sensors include displacement sensors, stress sensors, rain gauges, ground-based radars, and space-based radars;

[0008] Preprocess the data collected by various slope monitoring sensors to obtain time series of slope displacement, time series of slope stress, time series of rainfall, time queues of local deformation rates based on ground-based radar, and time queues of local deformation rates based on space-based radar;

[0009] Fuse the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queues of local deformation rates based on ground-based radar, and the time queues of local deformation rates based on space-based radar to obtain a multi-sensor fusion representation;

[0010] Based on the multi-sensor fusion representation, determine whether to trigger a warning signal.

[0011] According to another aspect of the present application, there is provided a multi-sensor based slope monitoring data fusion system, which includes:

[0012] A slope monitoring data acquisition module for acquiring data collected by various slope monitoring sensors, where the various slope monitoring sensors include displacement sensors, stress sensors, rain gauges, ground-based radars, and space-based radars;

[0013] A data preprocessing module for preprocessing the data collected by various slope monitoring sensors to obtain time series of slope displacement, time series of slope stress, time series of rainfall, time queues of local deformation rates based on ground-based radar, and time queues of local deformation rates based on space-based radar;

[0014] A multi-sensor data fusion module for fusing the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queues of local deformation rates based on ground-based radar, and the time queues of local deformation rates based on space-based radar to obtain a multi-sensor fusion representation;

[0015] A warning module for determining whether to trigger a warning signal based on the multi-sensor fusion representation.

[0016] Compared with the prior art, the multi-sensor-based slope monitoring data fusion method and system provided by the present application utilize a variety of slope monitoring sensors to collect various physical parameters of the slope, further construct the evidence bodies of each physical parameter, determine the basic probability assignment of each evidence body to different propositions by combining statistical analysis and expert experience, then gradually fuse all the evidence bodies using the D-S evidence synthesis rule to obtain a multi-sensor fusion representation, and further based on the multi-sensor fusion representation, according to the maximum confidence decision rule, determine the final stability state of the slope and take corresponding warning measures. In this way, multi-source heterogeneous data can be more comprehensively utilized, improving the accuracy and comprehensiveness of slope stability assessment and providing a strong guarantee for project safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] By describing the embodiments of the present application in more detail with reference to the accompanying drawings, the above and other objects, features, and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 It is a flowchart of the multi-sensor-based slope monitoring data fusion method according to an embodiment of the present application.

[0019] Figure 2 It is a schematic diagram of data flow of the multi-sensor-based slope monitoring data fusion method according to an embodiment of the present application.

[0020] Figure 3 It is a flowchart of step S3 in the multi-sensor-based slope monitoring data fusion method according to Embodiment 1 of the present application.

[0021] Figure 4 It is a flowchart of step S3 in the multi-sensor-based slope monitoring data fusion method according to Embodiment 2 of the present application.

[0022] Figure 5 It is a flowchart of step S32B in the multi-sensor-based slope monitoring data fusion method according to Embodiment 2 of the present application.

[0023] Figure 6 It is a block diagram of the multi-sensor-based slope monitoring data fusion system according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] Next, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0025] Example 1

[0026] Figure 1 It is a flowchart of a multi-sensor-based slope monitoring data fusion method according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of a multi-sensor-based slope monitoring data fusion method according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the multi-sensor-based slope monitoring data fusion method according to an embodiment of the present application includes the steps of: S1, obtaining data collected by a variety of slope monitoring sensors, the variety of slope monitoring sensors including displacement sensors, stress sensors, rain gauges, ground-based radars, and space-based radars; S2, preprocessing the data collected by the variety of slope monitoring sensors to obtain a time series of slope displacement, a time series of slope stress, a time series of rainfall, a time queue of local deformation rate based on the ground-based radar, and a time queue of local deformation rate based on the space-based radar; S3, fusing the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rate based on the ground-based radar, and the time queue of local deformation rate based on the space-based radar to obtain a multi-sensor fusion representation; S4, based on the multi-sensor fusion representation, determining whether to trigger an early warning signal.

[0027] In the above multi-sensor based slope monitoring data fusion method, in step S1, data collected by a variety of slope monitoring sensors is obtained. The variety of slope monitoring sensors includes displacement sensors, stress sensors, rain gauges, ground-based radars, and space-based radars. It should be understood that different types of sensors can reflect the physical characteristics of the slope from different perspectives. Specifically, displacement sensors can be used to monitor the displacement changes of the slope. By continuously recording the displacement data, it can intuitively reflect whether the slope is moving and the amplitude and direction of the movement, providing a key kinematic basis for judging the stability of the slope. Stress sensors can monitor the stress state inside the slope. Abnormal changes in the internal stress of the slope are often important signals indicating that its stability is threatened. For example, stress concentration may indicate a potential landslide risk. Rain gauges are used to record the rainfall in real time. Rainfall is one of the common external factors that trigger slope instability. Heavy rainfall will increase the weight and pore water pressure of the soil, reduce the shear strength of the soil, and thus affect the slope stability. Ground-based radars, with their high resolution and real-time monitoring capabilities, can accurately obtain the deformation rate information of local areas of the slope and promptly detect minor deformations on the slope surface. Space-based radars, from a macroscopic perspective, monitor the overall large area of the slope and provide information on the changes in the overall contour of the slope, helping to grasp the overall stability trend of the slope. By comprehensively obtaining data from a variety of sensors in this application, various physical parameters of the slope can be comprehensively grasped, thereby providing a rich and comprehensive information basis for the stability assessment of the slope.

[0028] In the above multi-sensor-based slope monitoring data fusion method, in step S2, the data collected by various slope monitoring sensors is preprocessed to obtain the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rates based on ground-based radar, and the time queue of local deformation rates based on space-based radar. It should be understood that in the actual monitoring process, various problems inevitably exist in the original collected data. On the one hand, factors such as noise interference in the measurement environment and the errors of the sensors themselves will cause a large amount of useless or even misleading information to be mixed into the data. For example, the displacement sensor may be affected by electromagnetic interference during the measurement process, resulting in fluctuations and deviations in the measurement data; the radar will also be affected by factors such as the atmospheric environment during the signal transmission and reception processes, generating noise signals. On the other hand, the data of different sensors may be inconsistent in time and space. For example, the sampling frequencies of the sensors are different, and the spatial reference systems of the data may also be different. Therefore, the present application further preprocesses the data collected by various slope monitoring sensors through preprocessing operations such as data cleaning, denoising, and normalization to improve the accuracy and consistency of the data. Specifically, for numerical data, taking the displacement sensor data as an example, filtering algorithms such as mean filtering and Kalman filtering can be used to remove the measurement noise, making the data more smoothly and accurately reflect the actual displacement changes; for radar data, through geometric correction and resolution enhancement and other processes, the spatial position and resolution of the data can be adjusted, so that the radar data at each time point is consistent and comparable in space. Then, by calculating the deformation amount between the corresponding pixel points in the radar data at adjacent time points and dividing the deformation amount by the time interval, the local deformation rate can be calculated. At the same time, the preprocessed data is organized into a time series form, forming the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rates based on ground-based radar, and the time queue of local deformation rates based on space-based radar, so as to intuitively observe the change trends of each parameter over time and provide an information basis in the time dimension for the evaluation and early warning of the slope state.

[0029] In the above multi-sensor-based slope monitoring data fusion method, in step S3, the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rates based on ground-based radar, and the time queue of local deformation rates based on space-based radar are fused to obtain a multi-sensor fusion representation. It should be understood that since the monitoring data of a single sensor often has limitations and cannot comprehensively reflect the stability status of the slope, it is necessary to fuse the data of multiple sensors to obtain more accurate and comprehensive slope stability information.

[0030] Figure 3It is a flowchart of step S3 in the multi-sensor-based slope monitoring data fusion method according to Embodiment 1 of the present application. As Figure 3 shown, the step S3 includes: S31A, respectively constructing evidential bodies of the time series of the slope displacement, the time series of the slope stress, the time series of the rainfall, the time queue of the local deformation rate based on the ground-based radar, and the time queue of the local deformation rate based on the space-based radar to obtain the first to fifth evidential bodies, wherein each of the evidential bodies contains one or more propositions; S32A, assigning probabilities to each proposition of each of the first to fifth evidential bodies based on statistical analysis and expert experience to obtain the first to fifth probability assignment evidential bodies; S33A, fusing the first to fifth probability assignment evidential bodies based on the D-S evidence theory to obtain a fused evidential body as the multi-sensor fusion representation.

[0031] Specifically, in order to establish a direct connection between the time series data collected by each sensor and the slope stability state, first, in the step S31A, based on the understanding of the relationship between the changes of different physical parameters and the slope stability, evidential bodies of the time series of the slope displacement, stress, rainfall, the local deformation rate based on the ground-based radar, and the local deformation rate based on the space-based radar are respectively constructed to obtain the first to fifth evidential bodies. Each of the evidential bodies contains one or more propositions. Here, the proposition is a different description of the slope stability state, such as "the slope is stable", "the slope is slightly unstable", "the slope is severely unstable", etc., which constitutes the identification framework of the evidential body. Specifically, taking the construction of the evidential body based on the time series of the slope displacement as an example, if the slope displacement continuously increases and exceeds a certain threshold within a period of time, according to the relevant theory of slope stability, it can be constructed as an evidential body about the "unstable" state of the slope. In this way, various types of time series data are transformed into evidential bodies with clear slope stability directions, providing a structured data basis for subsequent analysis, making the contribution of each sensor data to the slope stability assessment clearer and more definite, so as to facilitate the comprehensive consideration of multi-source information.

[0032] Since only constructing the evidential body cannot accurately measure the credibility of each proposition, therefore, the present application further determines the basic probability assignment (BPA) function of each evidential body for each proposition based on statistical analysis and expert experience, and quantifies the support degree of each evidential body for different propositions by assigning probabilities, so as to more accurately reflect the possibility of the slope being in different stability states. On the one hand, statistical analysis can dig out the probability relationship between the values of different physical parameters and the slope stability state from a large amount of historical data; on the other hand, expert experience takes into account factors such as actual geological conditions and slope characteristics that are difficult to fully reflected by data, and corrects and supplements the probability assignment.

[0033] In the specific implementation process of step S32A, taking the displacement sensor data as an example, first collect the stability statistical data of similar slopes in the area in the past under different displacement changes, and analyze the probabilities of the slope being in various states such as "stable" and "unstable" when the displacement increment is in different intervals. At the same time, combined with the empirical judgment of geological experts on the relationship between the slope stability and displacement in this area, determine the basic probability assignment (BPA) values of the displacement sensor evidence body for each proposition. Combining the two, determine the BPA value of "the slope is unstable" and the BPA value of "the slope is stable" when the displacement increment is in a specific range, such as 5 - 10 mm. Similarly, for other evidence bodies, the stress sensor determines its BPA function according to the relationship between the stress magnitude and the slope failure criterion, and the radar data determines its BPA function according to the correlation between the deformation characteristics and the stability, etc., so that the BPA value can reasonably reflect the support degree of each evidence body for different propositions. In this way, reasonable probability values are assigned to each proposition of each evidence body, forming the first to fifth probability assignment evidence bodies, so that the support strength of each evidence body for the slope stability assessment can be quantified, providing more comparable and operable data for the subsequent multi-source data fusion analysis.

[0034] Considering that in the process of evidence body fusion, it is inevitable to encounter the situation of evidence conflict, that is, the support degrees of different evidence bodies for the same proposition are quite different. For example, one evidence body strongly supports "the slope is stable", while another evidence body strongly supports "the slope is unstable". To solve this problem, in step S33A of this application, the Dempster - Shafer (D - S) evidence theory is adopted for the fusion of evidence bodies to solve the possible conflicts and uncertainties between evidence bodies. The D - S evidence theory is an effective method for dealing with uncertain information, and its principle is to combine multiple evidences by defining the basic probability assignment function, belief function and plausibility function. In this solution, the D - S evidence theory allows reasonable fusion of information in the case where the support degrees of different evidence bodies for different propositions are different or even in conflict, avoiding the one - sidedness of a single evidence.

[0035] In specific implementation, first, the conflict coefficient between each probability assignment evidence body is calculated to measure the degree of conflict between the evidences. When the conflict coefficient is large, it indicates that there are significant differences between the evidences. At this time, a credibility factor is introduced, and a credibility weight is assigned to each evidence body according to factors such as the accuracy, reliability of the sensor, and the accuracy of historical data. For example, if the displacement sensor has high accuracy and good stability, a higher credibility weight is assigned to it; if the stress sensor has relatively lower accuracy, a lower credibility weight is assigned. Then, the BPA value of the evidence body is corrected according to the credibility weight to reduce the negative impact of conflicting evidences on the fusion result while retaining useful information. After the conflict handling is completed, the Dempster combination rule of the D-S evidence theory is used for fusion. First, the fusion results of two-by-two evidence bodies are calculated. For example, first, the displacement evidence body and the stress evidence body are fused. For the proposition of "slope instability", the fused BPA value is calculated according to a specific formula of the Dempster combination rule. Then, the fusion result is used as a new evidence body and continues to be fused with the next evidence body (such as the rain gauge evidence body), and so on until all evidence bodies are fused. Finally, a comprehensive basic probability assignment function regarding the slope stability state, that is, the fused evidence body, is obtained. The fused evidence body synthesizes the information of multiple sensors, comprehensively reflects the stability state of the slope, and taking it as a multi-sensor fusion representation can effectively improve the utilization value of data and the reliability of the analysis result, greatly improve the accuracy and reliability of slope stability assessment, and effectively reduce the uncertainty of slope disaster risk assessment.

[0036] In the above multi-sensor-based slope monitoring data fusion method, in step S4, based on the multi-sensor fusion representation, it is determined whether to trigger an early warning signal. In an embodiment of the present application, based on the assignment probabilities of each proposition in the fused evidence body and adopting the maximum confidence decision principle, it is determined whether to trigger an early warning signal. That is, adopting the maximum confidence decision principle, based on the fused evidence body, the basic probability assignment values of each proposition related to slope stability are compared to determine whether to trigger an early warning signal. If the assignment probability of the proposition "the slope is severely unstable" in the fused evidence body is the largest and this assignment probability exceeds the preset danger threshold, it is determined that the slope is in the corresponding unstable state, and it is determined to trigger an early warning signal. Through various means such as text messages, sirens, and pop-ups on the monitoring system, early warning information is sent to relevant management departments and personnel to ensure that they can timely learn about the dangerous situation of the slope. At the same time, the support of each evidence body for different propositions can be further combined to deeply analyze the possible factors leading to slope instability. For example, if a relatively high degree of support for the proposition "rainfall causes slope instability" is found in the fusion result, and at the same time, abnormal changes also occur in displacement and stress data, through comprehensive analysis, it can be judged that it may be the recent heavy rainfall that causes the increase in soil saturation of the slope, thereby increasing the soil weight and reducing the shear strength, ultimately leading to changes in displacement and stress, resulting in the slope being in an unstable state. This provides a scientific and accurate decision-making basis for subsequent targeted prevention and control measures. The relevant departments can, based on the analysis results, quickly formulate and implement corresponding prevention and control measures such as drainage, reinforcement, and unloading, timely eliminate or reduce the slope disaster risk, and ensure the safety of the lives of people around the slope and the normal operation of various facilities.

[0037] In a specific example of the present application, a certain open-pit mine slope in Ordos is selected as the monitoring object. The slope is about 200 meters long, 50 meters high, with a slope of 45°, the surrounding geological conditions are complex, and it is greatly affected by seasonal rainfall.

[0038] A number of monitoring sensors are arranged on the slope, including 5 displacement sensors (model XX-100, measurement accuracy of ±0.5 mm), 3 stress sensors (model YY-50, measurement accuracy of ±0.1 MPa), 1 rain gauge (accuracy of 0.1 mm), and 1 set of ground-based radar and 1 set of space-based radar each. The displacement sensors and stress sensors collect data every 1 hour, the rain gauge collects rainfall data in real time, the ground-based radar conducts a scanning monitoring every 3 hours, and the space-based radar conducts monitoring when passing by every day.

[0039] For the displacement sensor data, the mean filter algorithm is used to remove noise with a window size of 5 data points; for the stress sensor data, linear interpolation is performed to repair a small amount of missing data; for the rain gauge data, outliers are removed (such as data with suddenly excessive rainfall that does not conform to local meteorological laws); for the ground-based radar data, geometric correction and resolution enhancement are carried out to make its spatial resolution reach 0.5 meters; for the space-based radar data, radiometric correction and geocoding are performed, and the preprocessed data is sorted into a time series format and stored in the database.

[0040] In the process of constructing the body of evidence, for the displacement sensor, if the displacement increment exceeds 5 mm within 3 consecutive measurement periods, an evidence body of "slope instability" is constructed, and the identification framework is {"slope stability", "slope instability"}; for the stress sensor, when the measured stress exceeds 80% of the design allowable stress, an evidence body of "potential slope instability" is constructed, and the identification framework is {"slope stability", "potential slope instability", "slope instability"}; for the rain gauge, when the rainfall exceeds 50 mm within 24 hours, an evidence body of "rainfall-induced slope instability" is constructed, and the identification framework is {"slope stability", "rainfall has an impact on the slope", "rainfall causes slope instability"}; for the ground-based radar, if the deformation rate of a local area on the slope surface is monitored to exceed 3 mm / d, an evidence body of "local slope instability" is constructed, and the identification framework is {"slope stability", "local slope instability", "overall slope instability"}; for the space-based radar, if a significant change in the overall contour of the slope is found in two consecutive observations, an evidence body of "overall slope deformation" is constructed, and the identification framework is {"slope stability", "slight slope deformation", "severe slope deformation"}.

[0041] Next, for the displacement sensor evidence body, based on the stability statistical data of similar slopes in the area under different displacement changes in the past and combined with the experience of geological experts, it is determined that when the displacement increment is between 5 - 10 mm, the BPA value of "slope is unstable" is 0.6, and the BPA value of "slope is stable" is 0.4; for the stress sensor evidence body, when the stress exceeds 80% of the design allowable stress, the BPA value of "slope is potentially unstable" is 0.5, the BPA value of "slope is stable" is 0.3, and the BPA value of "slope is unstable" is 0.2; for the rain gauge evidence body, when the rainfall exceeds 50 mm within 24 hours, the BPA value of "rainfall causes slope instability" is 0.4, the BPA value of "rainfall has an impact on the slope" is 0.4, and the BPA value of "slope is stable" is 0.2; for the ground-based radar evidence body, when the local deformation rate is between 3 - 5 mm / d, the BPA value of "local slope instability" is 0.7, the BPA value of "slope is stable" is 0.2, and the BPA value of "overall slope instability" is 0.1; for the space-based radar evidence body, when there are obvious changes in the overall contour of the slope, the BPA value of "severe slope deformation" is 0.5, the BPA value of "slight slope deformation" is 0.3, and the BPA value of "slope is stable" is 0.2. In this way, the BPA functions of each evidence body for different propositions are determined.

[0042] During the process of evidence conflict handling, first, calculate the conflict coefficients between each evidence body. For example, the conflict coefficient between the displacement sensor evidence body and the stress sensor evidence body is 0.3 (obtained through a specific conflict coefficient calculation method), and it is found that there is a relatively large conflict. According to factors such as the accuracy and reliability of the sensors, a credibility weight of 0.7 is assigned to the displacement sensor and a credibility weight of 0.6 is assigned to the stress sensor. The BPA values of the displacement sensor evidence body are corrected. For example, the corrected BPA value of "slope is unstable" is 0.6 × 0.7 = 0.42, and the corrected BPA value of "slope is stable" is 0.4 × 0.7 = 0.28; similar corrections are made to the BPA values of the stress sensor evidence body, and then subsequent fusion operations are carried out to reduce the adverse effects of conflicting evidence.

[0043] Next, fuse the displacement sensor evidence body and the stress sensor evidence body after conflict handling, and use the Dempster combination rule to calculate the fused BPA values. For example, for the proposition of "slope is unstable", the fused BPA value is calculated through a specific combination formula (the specific formula is based on the combination rule of D - S evidence theory), and then this fusion result is fused with the rain gauge evidence body, and so on, until all evidence bodies are fused to obtain the comprehensive BPA function regarding the slope stability state.

[0044] Finally, according to the fused comprehensive BPA function, the maximum confidence decision rule is adopted. If the BPA value of the proposition "the slope is seriously unstable" is the largest and exceeds the preset danger threshold of 0.6, it is determined that the slope is in a seriously unstable state. At this time, the system immediately sends early warning information to relevant management departments and personnel through text messages, sirens, etc. At the same time, by combining the support of each evidence body for different propositions, it is analyzed that it may be due to the increase in the soil saturation of the slope caused by recent heavy rainfall, which in turn causes changes in displacement and stress, and finally leads to the seriously unstable state of the slope, providing a decision-making basis for subsequent prevention and control measures such as drainage and reinforcement.

[0045] In summary, the multi-sensor based slope monitoring data fusion method according to the embodiments of the present application is elucidated. It uses a variety of slope monitoring sensors to collect various physical parameters of the slope, further constructs evidence bodies for each physical parameter, determines the basic probability assignment of each evidence body to different propositions by combining statistical analysis and expert experience, then uses the D-S evidence synthesis rule to gradually fuse all evidence bodies to obtain a multi-sensor fusion representation, and further based on the multi-sensor fusion representation, according to the maximum confidence decision rule, determines the final stability state of the slope and takes corresponding early warning measures. In this way, multi-source heterogeneous data can be more comprehensively utilized, the accuracy and comprehensiveness of slope stability assessment can be improved, and strong guarantee for project safety can be provided.

[0046] Embodiment 2

[0047] In particular, considering that in the field of slope monitoring, although the traditional multi-sensor data fusion method based on the D-S evidence theory can handle the uncertain information between different sensor data, its assignment of the BPA function of each evidence body depends on a large amount of prior knowledge and expert experience, which often has subjectivity and uncertainty in practical applications. To overcome this limitation, this embodiment proposes a multi-sensor data fusion method based on deep learning. It extracts the temporal change features of each sensor data and uses a neural network model to learn the temporal correlation and dependence relationship between various sensor data to achieve the temporal context joint perception of multi-sensor data, and then uses a classification algorithm to automatically identify the slope instability risk and give early warning prompts to reduce the dependence on prior knowledge and expert experience and improve the accuracy and reliability of slope monitoring.

[0048] Figure 4 It is a flowchart of step S3 in the multi-sensor based slope monitoring data fusion method according to Embodiment 2 of the present application. As Figure 4As shown, step S3 includes: S31B. Perform time series analysis on the time series of the slope displacement, the time series of the slope stress, the time series of the rainfall, the time queue of the local deformation rate based on the ground-based radar, and the time queue of the local deformation rate based on the space-based radar respectively to obtain the slope displacement time series correlation feature implicit coding vector, the slope stress time series correlation feature implicit coding vector, the rainfall time series correlation feature implicit coding vector, the first local deformation rate time series correlation feature implicit coding vector, and the second local deformation rate time series correlation feature implicit coding vector; S32B. Perform dynamic walk coding between sensor data based on causal triggering on the slope displacement time series correlation feature implicit coding vector, the slope stress time series correlation feature implicit coding vector, the rainfall time series correlation feature implicit coding vector, the first local deformation rate time series correlation feature implicit coding vector, and the second local deformation rate time series correlation feature implicit coding vector to obtain a multi-sensor data time series fusion coding feature vector as the multi-sensor fusion representation.

[0049] Specifically, in step S31B, in order to identify the time series change patterns and rules of each sensor data, the present application adopts a deep learning-based time series analysis technology to perform time series coding on the slope displacement, stress, rainfall, local deformation rate based on the ground-based radar, and local deformation rate based on the space-based radar respectively to capture the trends and periodic features of each sensor data changing over time. In the embodiment of the present application, the LSTM (Long Short-Term Memory) model is used as the core model for time series analysis to model the time series change characteristics of each sensor data. The LSTM model can effectively capture the long-term dependence relationships in the time series data by introducing mechanisms such as the forget gate, input gate, and output gate, and learn the time series change patterns of each sensor data within a long time range, thereby generating the slope displacement time series correlation feature implicit coding vector, the slope stress time series correlation feature implicit coding vector, the rainfall time series correlation feature implicit coding vector, the first local deformation rate time series correlation feature implicit coding vector, and the second local deformation rate time series correlation feature implicit coding vector, providing richer and more effective information for subsequent data fusion and classification tasks.

[0050] Specifically, in the step S32B, in order to effectively fuse the temporal variation information of each sensor data, the present application proposes a dynamic roaming encoding method between sensor data based on causal triggering. Based on the dynamic roaming mechanism in the graph neural network, by mining the temporal correlation dependencies between each sensor data, it simulates the causal relationship and temporal interaction between different sensor data, and performs multi-level temporal context joint perception and fusion on the implicit encoding vectors of the slope displacement temporal correlation features, the implicit encoding vectors of the slope stress temporal correlation features, the implicit encoding vectors of the rainfall temporal correlation features, the implicit encoding vectors of the first local deformation rate temporal correlation features, and the implicit encoding vectors of the second local deformation rate temporal correlation features, so as to learn a more comprehensive and accurate representation of the slope stability state, that is, the multi-sensor data temporal fusion encoding feature vector.

[0051] Figure 5 It is a flowchart of step S32B in the multi-sensor-based slope monitoring data fusion method according to Embodiment 2 of the present application. As Figure 5 shown, the step S32B includes: S321, arranging and combining the implicit encoding vectors of the slope displacement temporal correlation features, the implicit encoding vectors of the slope stress temporal correlation features, the implicit encoding vectors of the rainfall temporal correlation features, the implicit encoding vectors of the first local deformation rate temporal correlation features, and the implicit encoding vectors of the second local deformation rate temporal correlation features into a set of implicit encoding vectors of sensor data temporal correlation features; S322, performing implicit feature mining on each implicit encoding vector of sensor data temporal correlation features in the set of implicit encoding vectors of sensor data temporal correlation features to obtain a set of deep implicit temporal correlation feature encoding vectors of sensor data; S323, extracting the temporal semantic causal correlation topological features in the set of deep implicit temporal correlation feature encoding vectors of sensor data to obtain a temporal semantic causal correlation topological feature matrix between sensor data; S324, respectively inputting the temporal semantic causal correlation topological feature matrix between sensor data, the set of implicit encoding vectors of sensor data temporal correlation features, the temporal semantic causal correlation topological feature matrix between sensor data, and the set of deep implicit temporal correlation feature encoding vectors of sensor data into a dynamic roaming encoder based on a graph convolutional neural network model to obtain a multi-sensor data surface temporal context dynamic roaming encoding vector and a multi-sensor data hidden layer temporal context dynamic roaming encoding vector; S325, fusing the multi-sensor data hidden layer temporal context dynamic roaming encoding vector and the multi-sensor data surface temporal context dynamic roaming encoding vector to obtain the multi-sensor data temporal fusion encoding feature vector.

[0052] In an embodiment of the present application, the step S322 is expressed by the formula:

[0053] O = {x1, x2,..., x i ,..., x n}

[0054] v i = Sigmoid[Conv 1×1 (x i )]

[0055] D = {v1, v2,..., v i ,..., v n}

[0056] Among them, O represents the set of implicit encoding vectors of the temporal correlation features of sensor data. v1, x2, x i and x n respectively represent the 1st, 2nd, ith, and nth implicit encoding vectors of the temporal correlation features of sensor data in the set of implicit encoding vectors of the temporal correlation features of sensor data. n represents the number of vectors in the set of implicit encoding vectors of the temporal correlation features of sensor data. Conv 1×1 (·) represents a 1×1 convolution operation. Sigmoid represents an activation function. D represents the set of deep implicit temporal correlation feature encoding vectors of sensor data. v1, v2, v i and v n respectively represent the deep implicit temporal correlation feature encoding vectors of sensor data corresponding to x1, x2, x i and x n

[0057] That is, through point convolution encoding and non-linear activation operations, the important information in each implicit encoding vector of the temporal correlation features of sensor data is efficiently compressed, redundant information is removed, in order to mine the deep implicit information of each temporal correlation feature of sensor data, generate a deeper temporal feature representation, and obtain the set of deep implicit temporal correlation feature encoding vectors of sensor data.

[0058] ​In an embodiment of the present application, the step S323 includes: First, perform temporal correlation encoding on any two sensor data depth implicit temporal correlation feature encoding vectors in the set of sensor data depth implicit temporal correlation feature encoding vectors to obtain a set of temporal correlation encoding matrices between sensor data; Then, based on the statistical features of each temporal correlation encoding matrix between sensor data in the set of temporal correlation encoding matrices between sensor data, calculate its corresponding temporal semantic causal correlation factor between sensor data to obtain a temporal semantic causal correlation topology matrix composed of multiple temporal semantic causal correlation factors between sensor data, where the statistical features include the feature variance, maximum eigenvalue, feature mean, and causal correlation energy bias term of the temporal correlation encoding matrix between sensor data. The above calculation process can be expressed by the formula:

[0059]

[0060] where, v j represents the j-th sensor data depth implicit temporal correlation feature encoding vector in the set of sensor data depth implicit temporal correlation feature encoding vectors, represents the vector multiplication operation, (·) T represents the transpose of the vector, M i-j represents the temporal correlation encoding matrix between sensor data corresponding to v i and v j , σ 2 represents the feature variance of the temporal correlation encoding matrix between sensor data, μ(·) represents the feature mean of the matrix, λ represents the causal correlation energy bias term, max(·) represents taking the maximum value, t i-j represents the temporal semantic causal correlation factor between sensor data corresponding to v i and v j .

[0061] Here, in order to reveal the temporal correlation and dependence relationship between each sensor data, such as the temporal co-variation between displacement sensor data and stress sensor data, the temporal lag relationship between rainfall data and slope displacement, etc., the present application further constructs a global fine-grained temporal correlation coding matrix between every two sensor data deep implicit temporal correlation feature coding vectors in the set of sensor data deep implicit temporal correlation feature coding vectors, and uses a causal correlation energy metric function to perform feature distribution analysis on this temporal correlation coding matrix, so as to realize the explicit quantitative coding expression of the causal correlation between the corresponding two sensor data temporal features, and obtain a sensor data temporal semantic causal correlation topological matrix composed of temporal semantic causal correlation factors between multiple sensor data. Through the sensor data temporal semantic causal correlation topological matrix, the temporal dependence relationship and potential causal relationship between different sensor data can be revealed, which helps to more accurately understand the associated evolution process of the slope stability state and provides effective guidance for subsequent multi-sensor data fusion.

[0062] In a preferred example of the present application, if the characteristic variance of the temporal correlation coding matrix between the sensor data is greater than or equal to a preset threshold, calculate the average weighted value of the Euclidean distances between every two sensor data deep implicit temporal correlation feature coding vectors in the set of sensor data deep implicit temporal correlation feature coding vectors as the causal correlation energy bias term; if the characteristic variance of the temporal correlation coding matrix between the sensor data is less than the preset threshold, calculate the weighted value of the characteristic mean of the temporal correlation coding matrix between the sensor data as the causal correlation energy bias term, which is expressed by the formula:

[0063]

[0064] where ε represents a predetermined threshold, d(·,·) represents the Euclidean distance metric function, α and β respectively represent different weight parameters, L is the length of D, that is, the number of vectors in the set of sensor data deep implicit temporal correlation feature coding vectors,

[0065] That is, by treating the low-level causal associations in the temporal sequence correlation coding matrix of the sensor data as molecular-level relationships inferred based on statistical correlations, it is possible to further perform intervention prediction of causal association energy on the basis of a global fine-grained statistical association representation, and to study the causal association fine-grained structure and its dynamic regulation of the temporal sequence correlation coding matrix of the sensor data in terms of a high-dimensional and heterogeneous representation based on causal relationship omics. Among them, when the aggregative distribution representation of the causal graph is greater than a predetermined threshold, a bias occurs during source data integration based on the matrix representation of the temporal semantic causal association factor of the graph nodes, and when the aggregative distribution representation of the causal graph is less than the predetermined threshold, a condensed structure modeling can be directly performed through feature pattern integration compression. In this way, not only can the causal association energy in the temporal sequence correlation coding matrix of the sensor data be encoded and described, but also the implicit causal intervention prediction results can be condensed, thereby obtaining a more efficient revelation of key causal associations.

[0066] Then, the temporal sequence semantic causal association topology matrix between the sensor data is input into a causal trigger network based on a gated activation function to obtain a temporal sequence semantic causal association topology feature matrix between the sensor data, which is expressed by the formula:

[0067]

[0068]

[0069] where T represents the temporal sequence semantic causal association topology matrix between the sensor data, and t 1-1 , t 1-n , t n-1 and t n-n respectively represent the temporal sequence semantic causal association factors between two corresponding sensor data depth implicit temporal sequence association feature coding vectors in the set of sensor data depth implicit temporal sequence association feature coding vectors, softmax(·) represents the normalized exponential function, f trigger (·) represents the causal trigger network, τ represents the gating threshold, and M represents the temporal sequence semantic causal association topology feature matrix between the sensor data.

[0070] Here, after obtaining the temporal sequence semantic causal association topology matrix between the sensor data, a causal trigger network is further used to perform dynamic modeling on it. By introducing a normalized exponential function and a dynamic gating mechanism, dynamic causal triggering is performed on the topology matrix representation, and key causal paths are discriminated in the dynamic context, so as to strengthen important associations and weaken noise interference, and a temporal sequence semantic causal association topology feature matrix between the sensor data is obtained.

[0071] In an embodiment of the present application, the step S324 and the step S325 are expressed by the formula:

[0072]

[0073] H final = γ·H surface +(1 - γ)·H hidden

[0074] Among them, GCN(·) represents a graph convolutional neural network, and H surface represents the surface - layer temporal context dynamic walk encoding vector of multi - sensor data, and H hidden represents the hidden - layer temporal context dynamic walk encoding vector of multi - sensor data, and γ represents the fusion weight parameter, and H final represents the temporal fusion encoding feature vector of multi - sensor data.

[0075] That is, based on the set of the temporal semantic causal association topological feature matrix between the sensor data and the implicit encoding vector of the temporal association features of the sensor data, a surface - layer temporal context joint perception graph structure between the sensor data is constructed, and based on the set of the temporal semantic causal association topological feature matrix between the sensor data and the implicit encoding vector of the temporal association features of the sensor data, a hidden - layer temporal context joint perception graph structure between the sensor data is constructed. Using the dynamic graph representation learning method of the graph convolutional neural network (GCN), by continuously iteratively updating the node representations in the graph structure, the feature representation of each node (i.e., each sensor data) can fully integrate the information of its neighbor nodes, so as to realize the hierarchical learning of the temporal context information of multi - sensor data, obtain the surface - layer temporal context dynamic walk encoding vector of multi - sensor data and the hidden - layer temporal context dynamic walk encoding vector of multi - sensor data, and realize the multi - dimensional and multi - level perception of the slope stability state.

[0076] Finally, the surface - layer temporal context dynamic walk encoding vector of the multi - sensor data and the hidden - layer temporal context dynamic walk encoding vector of the multi - sensor data are fused by a weighted fusion method to comprehensively consider the temporal context information of multi - sensor data at different levels, generate the temporal fusion encoding feature vector of multi - sensor data, and use it as the multi - sensor fusion representation. Here, the temporal fusion encoding feature vector of the multi - sensor data synthesizes the temporal change features and potential causal association dependencies of each sensor data, and can more comprehensively and accurately reflect the stability state of the slope.

[0077] In step S4 of the multi-sensor-based slope monitoring data fusion method according to Embodiment 2 of the present application, determining whether to trigger a warning signal based on the multi-sensor fusion representation includes: inputting the multi-sensor data time-series fusion encoded feature vector into a slope state analyzer based on a classifier to obtain an analysis result, and the analysis result is used to indicate whether it is determined to trigger a warning signal. Specifically, the classifier can adopt machine learning algorithms such as support vector machine (SVM). During the training process, historical slope monitoring data and known slope stability states are used as training samples, and the parameters of the classifier are adjusted through continuous iterative calculations so that it learns the mapping relationship between different sensor data features and slope states, and can accurately distinguish slope sensor data in different stability states. Furthermore, during the actual application process, the multi-sensor data time-series fusion encoded feature vector is input into the trained classifier. The classifier learns the sensor data features contained in the multi-sensor data time-series fusion encoded feature vector, and combines the classification mapping relationship learned during the training process to automatically identify the stability state of the current slope, so as to map the multi-sensor data time-series fusion encoded feature vector to predefined classification labels ("trigger warning signal" and "do not trigger warning signal"), thereby realizing the automatic identification and warning prompt of slope instability risks. If the result output by the classifier is "trigger warning signal", it indicates that there is a high risk of instability for the current slope, and the system immediately triggers a warning mechanism to send warning information to relevant management departments and personnel through a preset communication method, prompting them to take necessary prevention and control measures. On the contrary, if the result output by the classifier is "do not trigger warning signal", it indicates that the current slope is in a relatively stable state, but the time-series changes of various sensor data still need to be continuously monitored to prevent problems before they occur.

[0078] Furthermore, the present application also provides a multi-sensor-based slope monitoring data fusion system.

[0079] Figure 6 It is a block diagram of the multi-sensor-based slope monitoring data fusion system according to the embodiment of the present application. As Figure 6As shown in the figure, the multi-sensor-based slope monitoring data fusion system 100 according to an embodiment of the present application includes: a slope monitoring data acquisition module 110, configured to acquire data collected by a variety of slope monitoring sensors, where the variety of slope monitoring sensors include displacement sensors, stress sensors, rain gauges, ground-based radars, and space-based radars; a data preprocessing module 120, configured to preprocess the data collected by the variety of slope monitoring sensors to obtain a time series of slope displacements, a time series of slope stresses, a time series of rainfall amounts, a time queue of local deformation rates based on the ground-based radar, and a time queue of local deformation rates based on the space-based radar; a multi-sensor data fusion module 130, configured to fuse the time series of slope displacements, the time series of slope stresses, the time series of rainfall amounts, the time queue of local deformation rates based on the ground-based radar, and the time queue of local deformation rates based on the space-based radar to obtain a multi-sensor fusion representation; and an early warning module 140, configured to determine whether to trigger an early warning signal based on the multi-sensor fusion representation.

[0080] Here, those skilled in the art can understand that the specific operations of the various modules in the above multi-sensor-based slope monitoring data fusion system have been described in detail in the description of the above multi-sensor-based slope monitoring data fusion method, and therefore, the repeated description thereof will be omitted.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-sensor-based slope monitoring data fusion method, characterized in that, Including: Obtaining data collected by a variety of slope monitoring sensors, where the variety of slope monitoring sensors include displacement sensors, stress sensors, rain gauges, ground-based radars, and space-based radars; Preprocessing the data collected by the variety of slope monitoring sensors to obtain time series of slope displacement, time series of slope stress, time series of rainfall, time queue of local deformation rate based on ground-based radar, and time queue of local deformation rate based on space-based radar; Fusing the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rate based on ground-based radar, and the time queue of local deformation rate based on space-based radar to obtain a multi-sensor fusion representation; Based on the multi-sensor fusion representation, determining whether to trigger a warning signal.

2. The multi-sensor-based slope monitoring data fusion method according to claim 1, wherein, Fusing the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rate based on ground-based radar, and the time queue of local deformation rate based on space-based radar to obtain a multi-sensor fusion representation, including: Respectively constructing evidence bodies for the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rate based on ground-based radar, and the time queue of local deformation rate based on space-based radar to obtain the first to fifth evidence bodies, where each of the evidence bodies contains one or more propositions; Assigning probabilities to each proposition of each of the first to fifth evidence bodies based on statistical analysis and expert experience to obtain the first to fifth probability assignment evidence bodies; Fusing the first to fifth probability assignment evidence bodies based on the D-S evidence theory to obtain a fused evidence body as the multi-sensor fusion representation.

3. The multi-sensor based slope monitoring data fusion method according to claim 2, wherein Based on the multi-sensor fusion representation, determining whether to trigger a warning signal, including: Based on the assigned probabilities of each proposition in the fused evidence body and adopting the maximum confidence decision principle, determining whether to trigger a warning signal.

4. The multi-sensor-based slope monitoring data fusion method according to claim 3, characterized in that, Based on the assigned probabilities of each proposition in the fused evidence body and adopting the maximum confidence decision principle, determining whether to trigger a warning signal, including: If the assigned probability of the proposition "the slope is severely unstable" in the fused evidence body is the largest and this proposition's assigned probability exceeds a preset danger threshold, determining to trigger a warning signal.

5. The multi-sensor based slope monitoring data fusion method according to claim 1, wherein Fusing the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rate based on ground-based radar, and the time queue of local deformation rate based on space-based radar to obtain a multi-sensor fusion representation, including: Perform time series analysis on the time series of the slope displacement, the time series of the slope stress, the time series of the rainfall, the time series queue of the local deformation rate based on ground-based radar, and the time series queue of the local deformation rate based on space-based radar to obtain the implicit encoding vector of the time series correlation characteristics of the slope displacement, the implicit encoding vector of the time series correlation characteristics of the slope stress, the implicit encoding vector of the time series correlation characteristics of the rainfall, the first implicit encoding vector of the time series correlation characteristics of the local deformation rate, and the second implicit encoding vector of the time series correlation characteristics of the local deformation rate; Perform dynamic random walk encoding between sensor data based on causal triggering on the implicit encoding vector of the time series correlation characteristics of the slope displacement, the implicit encoding vector of the time series correlation characteristics of the slope stress, the implicit encoding vector of the time series correlation characteristics of the rainfall, the first implicit encoding vector of the time series correlation characteristics of the local deformation rate, and the second implicit encoding vector of the time series correlation characteristics of the local deformation rate to obtain the time series fusion encoding feature vector of multi-sensor data as the multi-sensor fusion representation.

6. The multi-sensor-based slope monitoring data fusion method according to claim 5, wherein Performing dynamic random walk encoding between sensor data based on causal triggering on the implicit encoding vector of the time series correlation characteristics of the slope displacement, the implicit encoding vector of the time series correlation characteristics of the slope stress, the implicit encoding vector of the time series correlation characteristics of the rainfall, the first implicit encoding vector of the time series correlation characteristics of the local deformation rate, and the second implicit encoding vector of the time series correlation characteristics of the local deformation rate to obtain the time series fusion encoding feature vector of multi-sensor data as the multi-sensor fusion representation, including: Arrange and combine the implicit encoding vector of the time series correlation characteristics of the slope displacement, the implicit encoding vector of the time series correlation characteristics of the slope stress, the implicit encoding vector of the time series correlation characteristics of the rainfall, the first implicit encoding vector of the time series correlation characteristics of the local deformation rate, and the second implicit encoding vector of the time series correlation characteristics of the local deformation rate into a set of implicit encoding vectors of the time series correlation characteristics of sensor data; Perform implicit feature mining on each implicit encoding vector of the time series correlation characteristics of sensor data in the set of implicit encoding vectors of the time series correlation characteristics of sensor data to obtain a set of deep implicit time series correlation feature encoding vectors of sensor data; Extract the time series semantic causal association topological features in the set of deep implicit time series correlation feature encoding vectors of sensor data to obtain the time series semantic causal association topological feature matrix between sensor data; Input the time series semantic causal association topological feature matrix between sensor data, the set of implicit encoding vectors of the time series correlation characteristics of sensor data, the time series semantic causal association topological feature matrix between sensor data, and the set of deep implicit time series correlation feature encoding vectors of sensor data into the dynamic random walk encoder based on the graph convolutional neural network model respectively to obtain the surface layer time series context dynamic random walk encoding vector of multi-sensor data and the hidden layer time series context dynamic random walk encoding vector of multi-sensor data; Fuse the hidden layer time series context dynamic random walk encoding vector of multi-sensor data and the surface layer time series context dynamic random walk encoding vector of multi-sensor data to obtain the time series fusion encoding feature vector of multi-sensor data.

7. The multi-sensor-based slope monitoring data fusion method according to claim 6, wherein Extract the temporal semantic causal association topological features in the set of temporal implicit sequential association feature encoding vectors of the sensor data to obtain a temporal semantic causal association topological feature matrix between sensor data, including: Perform temporal association encoding on any two temporal implicit sequential association feature encoding vectors in the set of temporal implicit sequential association feature encoding vectors of the sensor data to obtain a set of temporal association encoding matrices between sensor data; Based on the statistical features of each temporal association encoding matrix in the set of temporal association encoding matrices between sensor data, calculate the corresponding temporal semantic causal association factors between sensor data to obtain a temporal semantic causal association topological matrix composed of multiple temporal semantic causal association factors between sensor data, where the statistical features include the feature variance, maximum eigenvalue, feature mean, and causal association energy bias term of the temporal association encoding matrix between sensor data; Input the temporal semantic causal association topological matrix between sensor data into a causal trigger network based on a gated activation function to obtain the temporal semantic causal association topological feature matrix between sensor data.

8. The multi-sensor-based slope monitoring data fusion method according to claim 7, wherein If the feature variance of the temporal association encoding matrix between sensor data is greater than or equal to a preset threshold, calculate the average weighted value of the Euclidean distances between every two temporal implicit sequential association feature encoding vectors in the set of temporal implicit sequential association feature encoding vectors of the sensor data as the causal association energy bias term; if the feature variance of the temporal association encoding matrix between sensor data is less than the preset threshold, calculate the weighted value of the feature mean of the temporal association encoding matrix between sensor data as the causal association energy bias term.

9. The multi-sensor-based slope monitoring data fusion method according to claim 8, characterized in that, Based on the multi-sensor fusion representation, determine whether to trigger a warning signal, including: inputting the multi-sensor data temporal fusion encoding feature vector into a slope state analyzer based on a classifier to obtain an analysis result, and the analysis result is used to indicate whether to determine to trigger a warning signal.

10. A slope monitoring data fusion system based on multi-sensors, characterized in that, Including: A slope monitoring data acquisition module, configured to acquire data collected by a variety of slope monitoring sensors, where the variety of slope monitoring sensors include displacement sensors, stress sensors, rain gauges, ground-based radars, and space-based radars; A data preprocessing module, configured to preprocess the data collected by the variety of slope monitoring sensors to obtain a time series of slope displacement, a time series of slope stress, a time series of rainfall, a time queue of local deformation rates based on ground-based radars, and a time queue of local deformation rates based on space-based radars; A multi-sensor data fusion module, configured to fuse the time series of slope displacement, the time series of slope stress, the time series of rainfall, the time queue of local deformation rates based on ground-based radars, and the time queue of local deformation rates based on space-based radars to obtain a multi-sensor fusion representation; A warning module, configured to determine whether to trigger a warning signal based on the multi-sensor fusion representation.

Images